forked from catalyst-team/catalyst
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconfig_dqn.yml
200 lines (178 loc) · 3.53 KB
/
config_dqn.yml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
args:
logdir: ./logs/rl-gym-dqn # change me
expdir: null
vis: 0
infer: 0 # change me
valid: 1 # change me
train: 1 # change me
db:
db: RedisDB
port: 12000
prefix: gym-dqn # TODO: remove
environment:
environment: GymEnvWrapper
env_name: LunarLander-v2
history_len: &history_len 3
frame_skip: 2
reward_scale: 1.0
agents:
critic:
agent: ActionCritic
state_net_params: # state -> hidden representation
observation_net_params:
_network_type: "linear"
history_len: *history_len
features: [128] # first hidden would be taken from state_shape
use_bias: false
normalization: LayerNorm
activation: ReLU
main_net_params:
features: [128, 128]
use_bias: false
normalization: LayerNorm
activation: ReLU
value_head_params: # hidden representation -> ~policy
in_features: 128 # out features would be taken from action_shape
# distribution: categorical
# num_atoms: 51
# values_range: [-300.0, 300.0]
# distribution: quantile
# num_atoms: 51
algorithm:
algorithm: DQN
n_step: 1
gamma: 0.99
critic_tau: 1.0
critic_loss_params:
criterion: HuberLoss
clip_delta: 10.0
critic_optimizer_params:
optimizer: Adam
lr: 0.0003
trainer:
batch_size: 256 # transitions
num_workers: 1
epoch_len: 500 # batches, 128k samples
replay_buffer_size: 1000000 # transitions
min_num_transitions: 5000 # transitions
save_period: 100 # epochs
weights_sync_period: 1 # epochs
target_update_period: 250 # batches, 64k updates
max_updates_per_sample: 32
sampler:
weights_sync_period: 1
exploration_params:
- exploration: Boltzmann
probability: 0.5
temp_init: 1.0
temp_final: 0.5
annealing_steps: 10000
- exploration: EpsilonGreedy
probability: 0.45
eps_init: 1.0
eps_final: 0.5
annealing_steps: 10000
- exploration: Greedy
probability: 0.05
valid_seeds: [
1608637542,
1273642419,
1935803228,
787846414,
996406378,
1201263687,
423734972,
415968276,
670094950,
1914837113,
669991378,
429389014,
249467210,
1972458954,
1572714583,
1433267572,
434285667,
613608295,
893664919,
648061058,
88409749,
242285876,
2018247425,
953477463,
1427830251,
1883569565,
911989541,
3344769,
780932287,
2114032571,
787716372,
504579232,
1306710475,
479546681,
106328085,
30349564,
1855189739,
99052376,
1250819632,
106406362,
480404538,
1717389822,
599121577,
200427519,
1254751707,
2034764475,
1573512143,
999745294,
1958805693,
389151677,
1224821422,
508464061,
857592370,
1642661739,
61136438,
2075460851,
396917567,
2004731384,
199502978,
1545932260,
461901618,
774414982,
732395540,
1934879560,
279394470,
56972561,
1927948675,
1899242072,
1999874363,
271820813,
1324556529,
1655351289,
1308306184,
68574553,
419498548,
991681409,
791274835,
1035196507,
1890440558,
787110843,
524150214,
472432043,
2126768636,
1431061255,
147697582,
744595490,
1758017741,
1679592528,
1111451555,
782698033,
698027879,
1096768899,
1338788865,
1826030589,
86191493,
893102645,
200619113,
290770691,
793943861,
134489564
]